Abstract
Recently, studies have reported that pan-viral serology signatures may be predictive for liver cancer development. However, whether these same findings are observed for prospective studies has not been previously investigated. The nested case-control analysis included 191 persons who developed liver cancer and 382 controls from the PLCO prospective cohort. The presence of circulating antibodies, measured by VirScan, was determined in serum samples obtained at study recruitment. The presence of antibodies was compared between cases and controls using multivariable conditional logistic regressions, and prediction models were used to estimate whether exposures predicted liver cancer development. No significant associations were found between antibodies to viruses, bacteria or allergens and liver cancer risk after adjustment for multiple testing. The agent most significantly associated with risk was hepatitis C virus (HCV), but it was only detected among 23 participants (odds ratio (OR): 3.98; 95% confidence intervals (CI):1.59-9.99; p = 0.0032, False Discovery Rate (FDR) = 0.35). In prediction models based on 109 antibody features, no associations with liver cancer risk were observed (area under the curve [AUC]: 0.52-0.54). In analyses restricted to the most common type of liver cancer, hepatocellular carcinoma, the association with HCV was stronger (OR: 23.16, 95% CI: 4.55-117.68; FDR p-value = 0.0016), although prediction models based on all detected antibodies were similar (AUC = 0.55; 95% CI:0.43-0.68). Antibodies to no infectious agents, other than HCV, were found to be prospectively associated with liver cancer risk. The utility of using an antibody exposure signature prospectively for liver cancer development needs to be further explored.